Latest in AI
Every brief, newest first. Each one links to its primary source.
-
OpenAI opens its Daybreak cyber programme to Ukraine
Ukraine’s government gets access to OpenAI’s gated Daybreak security programme, with no stated duration, cost or oversight terms.
-
Altman asks the UN Security Council for common frontier AI standards
Altman told the Security Council that countries need shared standards for measuring, with no treaty, body or timeline attached.
-
Claude agents flag an enzyme system with CRISPR-like repeats
Anthropic says ~950 Claude agents spent 21 hours finding a CRISPR-like repeat array in a phage. What the system does is still unknown.
-
A faster checkpoint for NVIDIA’s surgical video model
NVIDIA has added a distilled checkpoint to its open-weights surgical video model, claiming a near thirteenfold speed-up, with the card explicit that it is…
-
NVIDIA documents its open speech-to-speech model, weak spots included
The technical report for NVIDIA’s open full-duplex speech model reports strong interruption handling alongside noticeably weaker tool-calling accuracy.
-
Fifty models, 258 cognitive experiments, and a gap that is not closing fast
A new framework re-runs 258 cognitive science experiments on 50 language models and finds their agreement with human answers well short of how well…
-
Agents can copy how an app looks. Finishing it is another matter
A new benchmark gives the leading model 58.1% overall but only 2.8% of tasks passing every test, which says something about how agent scores…
-
Qwen’s new image model is open-weight, not open-licensed
Alibaba’s Qwen team has published a 7B image generation and editing model with native transparency, but the weights carry a research licence rather than…
-
South-East Asia builds sovereign AI on NVIDIA’s open Nemotron models
Singapore, Malaysia, Vietnam and Thailand are fine-tuning NVIDIA’s open weights for local languages rather than training from scratch.
-
Anthropic prices Opus 5.5 down, and prints the benchmarks it loses
Opus 5.5 costs 40% less to run than Opus 5, and Anthropic’s own table shows GPT-6 Astra ahead on two of its eight benchmarks.